| Age | Commit message (Collapse) | Author |
|---|
|
* siri/cover-tests: (21 commits)
[common_test] Extend timer for flushing error logger
[cover] Allow reconnection if node has been disconnected or down
[cover] Don't kill remote nodes when connection to main node is lost
[test_server] Add option {start_cover,false} to test_server:start_node
Use code:lib_dir instead of code:which to get application directory
[common_test] Add test for OTP-9956
Include all kernel modules in code coverage analysis
[common_test] Add test suite for code coverage support
[common_test, test_server] Don't flush cover if cover is not running
[common_test] Add option cover_stop
[test_server] Allow cross cover analysis when testing through common_test
[test_server] Start cover in test_server:wait_for_node
[test_server] Multiply timers with timetrap_scale_factor when starting nodes
Include all stdlib modules in code coverage analysis
[test_server] Include all test_server modules in code coverage analysis
Skip epp_SUITE:otp_8911 if cover is running
[common_test] Start cover on slave nodes if running cover tests
[common_test] Don't stop cover before stopping slave node
[test_server] Don't stop cover after test is finished
[cover] Add support for test_server
...
OTP-10427
|
|
* bjorn/ct/stability/OTP-10480:
Simplify managment of test_server_loc
Keep the information about the current test case consistent
test_server: Eliminate the Loc argument for do_end_tc_call()
Refactor the handling of the make_priv_dir message
test_server: Refactor run_test_case_msgloop()
|
|
Because of a forgotten "Fd", configuration information intended for
ct_master_log.html ended up being written to standard output.
|
|
* raimo/IPV6_V6ONLY/OTP-8928:
kernel: Document socket option ipv6_v6only
kernel: Add test cases for socket option ipv6_v6only
erts,kernel: Implement socket option ipv6_v6only in erlang code
erts: Implement socket option IPV6_V6ONLY
erts: Add configure test for IPV6_V6ONLY
|
|
|
|
|
|
|
|
|
|
* dgud/stdlib/proc_lib_hangs/OTP-9803:
stdlib: Monitor proc_lib:start'ed processes to avoid waiting forever
|
|
Monitor the spawned process in proc_lib:start/[3|4|5], so
that a proper error is returned, if the spawned process crashes
before proclib:init_ack/1 is sent.
Previously the calling process would hang forever or until specified
timeout. Start link catches these errors but start did not.
start now behaves as start_link if process traps exit.
|
|
* dgud/test-fixes:
test: fix timetraps
kernel: Extend test timeout for cover tests
stdlib: Fix test timeouts in cover runs
runtime_tools: Fix failing testcase
|
|
Use ct:timetrap/1 instead of test_server:timetrap/1 for really long
test_cases.
Commontest sets up a default timetrap on 30 min, which is not cancelled
if testserver:timetrap called.
|
|
|
|
|
|
|
|
OTP-10040
If a busy test case generated lots of error messages,
cth_log_redirect:post_end_per_testcase would crash with a timeout
while waiting for the error logger to finish handling all error
messages. The default timer is 5 seconds. This has now been extended
to 5 minutes.
|
|
OTP-10523
Earlier, if the connection to a remote cover node was lost, all cover
data was lost and the cover_server on the remote node would die. This
would cause problems if there were cover compiled modules that would
still be executed since they would attempt to write to the no longer
existing ets tables belonging to the cover_server.
This commit changes this behavior so that the cover_server on the
remote node will survive connection loss and continue collecting cover
data. If the connection is re-established then the main node will sync
with the remote node again and cover data will not be lost (unless the
node was down).
|
|
OTP-10523
Since the will probably be cover compiled modules left in the remote
node, we want to keep the cover_server and it's ets tables even if
connection to remote node is lost. This way it will not crash (due to
ets:insert in non existing table) if functions in cover compiled
modules are executed.
|
|
By default the test server will start cover on all nodes when the test
is run with code coverage analysis. To make sure cover is not started
on a new node, this option can be set.
This can be useful if the test itself starts cover or if the new node
for some reason can not run cover compiled code.
|
|
code_SUITE:ext_mod_dep used code:which(kernel) to get the path to the
kernel ebin directory. This failed when kernel was cover compiled,
since code:which then returns 'cover_compiled'. Using code:lib_dir
instead solved the problem.
|
|
OTP-9956
Test that the test suite itself can be cover compiled, and that
data_dir is still set correctly.
|
|
|
|
|
|
ct_slave, ct_test_support and test_server_test_lib always called
cover:flush/1 when stopping slave nodes. If cover was not running,
this would cause cover_server to be started and thus
test_server:is_cover/0 to return true afterwards. This has been
corrected.
|
|
By default, test_server will always stop cover (and thus load back the
non cover compiled original beam files) after a test with code
coverage analysis is completed. The new option allows the common_test
user to specify that cover shall not be stopped. This can be useful if
there are processes still running old code, i.e. processes that have
not done any fully qualified function call after the cover
compilation, since loading the original code then will kill those
processes. This is only recommended if the erlang node is to be
stopped after the test run, or if cover can be manually stopped.
|
|
The cross cover analysis functionality in test_server was broken since
OTP converted to running daily tests through common_test. This has
been corrected.
|
|
Without this change, there is no support in test_server for starting
cover on a node started with option {wait,false}.
|
|
In test_server_ctrl:start_node and wait_for_node, the timer in
controller_call was not multiplied by the timetrap_scale_factor. This
caused unexpected behaviour since the scale factor was used furhter
down the the code, and timeout would happen in unexpected order when
cover was running.
|
|
|
|
|
|
This is because this test case itself starts and stops cover.
|
|
Nodes started with ct_slave should automatically run cover when tests
are executed with cover analysis. Nodes must also be flushed for cover
data before stopped.
|
|
In order to avoid crash of processes that hold old code, cover is no
longer stopped before slave nodes are stopped. Instead cover:flush is
used which only fetches cover data from the slave.
This, along with similar changes in test_server allows all common_test
modules to be included in cover analysis.
|
|
This commit removes all calls to cover:stop from test_server. This is
to avoid crash in some processes due to old code.
The correction shall be further refined for general use. There will,
for instance, be an option to turn off cover:stop/0 after completed
test run, so the default behaviour is backwards compatible.
|
|
OTP-10523
* Added cover:flush(Nodes), which will fetch data from remote nodes
without stopping cover on those nodes.
* Added cover:get_main_node(), which returns the node name of the main
node. This is used by test_server to avoid {error,not_main_node}
when a slave starts another slave (e.g. in test_server's own tests).
|
|
|
|
Before line numbers were included in exceptions (in R15), there were
parse transforms and macros that would keep the test_server_loc
process dictionary variable updated; therefore there is a mod_loc/1
function that will normalize the representation of locations.
Simplify the code as following:
* 'test_server_loc' should always contain a list with one or more
tuples. ({M,F} or {M,F,Line})
* At the beginning of each test case, set 'test_server_loc' to
[{Module,Func}] of the the currently executing test case.
* Stop updating 'test_server_loc', except when an exception occurs.
(It used to be updated when running 'init_per_testcase' and so on.)
* Remove the mod_loc/1 function.
|
|
Three pieces of information could be out of sync in testcase
supervisor process at the time when a timetrap occurred:
* test_server_loc (process dictionary) - may indicate that a
framework function is executing
* test_server_init_or_end_conf (process dictionary) - indicates whether
init_per_testcase or end_per_testcase is executing
* The current configuration (#st.config) - set using a synchronous
call
There could be in a crash in spawn_fw_call/7 because the current
configuration was not defined when it was expected to.
To avoid the problem, introduce set_tc_state/2 (and a corresponding
message) to allow setting both an indication what the testcase is
executing (e.g. init_per_testcase, framework call, and so on), and
the current configuration. Use only that information to handle
a timetrap timeout (and aborted testcase and the other reasons
for the testcase process to terminate). Completely remove
test_server_init_or_end_conf.
|
|
The Loc argument was use to determine the name of the currently
executing test case, in case that the M and F arguments did not
specify a test case.
Since a previous commit makes sure that the M and F arguments *are*
valied, we can eliminate the Loc argument.
|
|
The code is not actually shorter, but it avoids duplicating the code
for producing an error tuple when theres is no priv_dir tuple in
the config data.
|
|
Keeping all state as separate function arguments does not make it
easy to add more pieces of data to the state. Case in point is the
CurrConf argument where up to three separate items have been
shoehorned in.
Instead of the arguments, introduce a state record to hold all of the
different pieces of state (taking care to separate the former CurrConf
variable into separate fields in the record).
While at it, fix a bug. The name of the currently executing test
case (module and function) used to be stored in a tuple in the CurrConf
variable, and it would only be available after 'set_curr_conf' message
has been received. Depending on timing, it was possible in rare
circumstances for an EXIT signal to arrive before the 'set_curr_conf'
message. Avoid this problem by putting the current test case
name into the record from the beginning. That also means that we
can eliminate the workaround of getting the currently executing test case
from the stack trace using the get_mf/1.
|
|
* bjorn/compiler/test-cases:
Correct typo in test suite name
compiler: Run testcases in parallel
|
|
* bjorn/test_server/test-cases:
test_server tests: Be kind to Windows by using shorter pathnames
|
|
* bjorn/erts/beam_makeops:
beam_makeops: Turn on warnings and eliminate existing warnings
beam_makeops: Eliminate a deprecation warning
|
|
* bjorn/ct/separate-io-server/OTP-10101:
Test ct:capture/start/stop/get
Introduce ct_group_leader_SUITE that does nasty things with group leaders
Eliminate unexpected I/O to the minor log file
Introduce test_server_io and test_server_gl
Clean up initialization of parallel test cases
Introduce is_io_buffered/0 to somewhat improve readability
test_server_ctrl: Consistently use set_io_buffering/1
|
|
* siri/test_server/faulty-data_dir-when-cover/OTP-9956:
[test_server] Set data_dir correctly when suite is cover compiled
|
|
If the test suite itself was included in code coverage analysis, then
test_server_ctrl would not manage to set data_dir correctly for the
test, since it relied on the result of code:which(Suite). This has
been corrected.
|
|
The pathnames gets too long for Windows if we use priv_dir as working
directory for the test_server tests in the slave node. Use data_dir
instead. Revert this commit when the run-time system can handle long
pathnames on Windows.
|
|
Conflicts:
erts/etc/common/heart.c
|
|
* egil/ensure-erl_crash.dump/OTP-10422:
test: Relax timeouts for heart_SUITE
erts: Fix crash dump write to port hack
erts: Fix lock check assertion
doc: Document ERL_CRASH_DUMP_SECONDS behaviour
test: Add test for heart restart on crash
test: Add test for heart restart on crash
erts: Change ERL_CRASH_DUMP_SECONDS behaviour
test: Refactor away ?line macros in heart_SUITE
erts: Search for heart in ports that are alive
heart: Refactor heart debugging
erts, heart: Ensure erl_crash.dump is written
|
